Temporal Multimodal Fraud Detection via Cross-Year Financial Statement Changes
International Journal of Business and Social Science Studies · 10.09.2026 · 6 (2) · 196-210
DOI: 10.66795/ijbsss72 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Business and Social Science Studies, online erschienen am 10.09.2026. Verfasst von Derek Wang und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Financial statement fraud is usually screened from a single year of reported numbers, even though misreporting unfolds over several fiscal years and must be absorbed at the same time by the balance sheet, the income statement, and the market valuation of the firm. This paper presents CYMF, a cross-year multimodal fusion framework that represents a firm’s recent reporting history as aligned yearly token sequences of three statement modalities (balance-sheet stocks, income and financing flows, and market valuation) together with an articulation stream of eighteen cross-statement consistency residuals. A temporal expert encodes these tokens with causal cross-year multimodal attention, learns a self-supervised expectation of every year’s changes from the preceding years, and turns the deviations from that expectation into an explicit evidence stream, in the spirit of the analytical procedures used by auditors. Gated evidence fusion weighs the modalities for each firm-year, and a validation-calibrated rank fusion combines the temporal expert with a boosted-tree expert on current-year levels. On the public AAER-labelled panel of 146,045 U.S. firm-years, evaluated out of sample over 2003–2008 with a two-year label gap and serial-fraud control, CYMF attains the highest mean AUC (0.724) and average precision (0.0205) among nine models, and its temporal expert alone raises NDCG@1% from 0.011 to 0.030 relative to the RUSBoost benchmark.
Zitieren
APA
Wang, D., & Bennett, S. (2026). Temporal Multimodal Fraud Detection via Cross-Year Financial Statement Changes. International Journal of Business and Social Science Studies, 6(2), 196-210. https://doi.org/10.66795/ijbsss72
Vancouver
Wang D, Bennett S. Temporal Multimodal Fraud Detection via Cross-Year Financial Statement Changes. International Journal of Business and Social Science Studies. 2026;6(2):196-210. doi:10.66795/ijbsss72
BibTeX
@article{Wang2026,
author = {Wang, Derek and Bennett, Sophie},
title = {Temporal Multimodal Fraud Detection via Cross-Year Financial Statement Changes},
journal = {International Journal of Business and Social Science Studies},
year = {2026},
volume = {6},
number = {2},
pages = {196--210},
doi = {10.66795/ijbsss72}
}
EUVIMED